Contents (Содержание)

1) Diving into data (Знакомство с данными)

2) Vizualization (Визуализация)

3) CatBoostClassifier

4) Economic effect (Экономический эффект)

import pandas as pd
import numpy as np
pd.set_option("display.max_columns", 500)

from pandas_profiling import ProfileReport

from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.metrics import accuracy_score, f1_score, roc_auc_score, classification_report
from sklearn.preprocessing import StandardScaler

from catboost import CatBoostClassifier

import warnings
warnings.filterwarnings("ignore")

import plotly.graph_objs as go

import matplotlib.pyplot as plt
import seaborn as sns
sns.set_style("darkgrid")

from plotly.offline import download_plotlyjs, init_notebook_mode, iplot
init_notebook_mode(connected=True)
#import plotly.express as px

1) Diving into data (Знакомство с данными)

# Author of this kaggle dataset had two synthetic features, so I dropped them before and made file "BankChurners_1.csv"
path = "/Users/salikh-khabibullin/Desktop/Pet_Project/BankChurn/"
data = pd.read_csv(path + "BankChurners_1.csv")
data.shape

(10127, 21)

data.head()

	CLIENTNUM	Attrition_Flag	Customer_Age	Gender	Dependent_count	Education_Level	Marital_Status	Income_Category	Card_Category	Months_on_book	Total_Relationship_Count	Months_Inactive_12_mon	Contacts_Count_12_mon	Credit_Limit	Total_Revolving_Bal	Avg_Open_To_Buy	Total_Amt_Chng_Q4_Q1	Total_Trans_Amt	Total_Trans_Ct	Total_Ct_Chng_Q4_Q1	Avg_Utilization_Ratio
0	768805383	Existing Customer	45	M	3	High School	Married	$60K - $80K	Blue	39	5	1	3	12691.0	777	11914.0	1.335	1144	42	1.625	0.061
1	818770008	Existing Customer	49	F	5	Graduate	Single	Less than $40K	Blue	44	6	1	2	8256.0	864	7392.0	1.541	1291	33	3.714	0.105
2	713982108	Existing Customer	51	M	3	Graduate	Married	$80K - $120K	Blue	36	4	1	0	3418.0	0	3418.0	2.594	1887	20	2.333	0.000
3	769911858	Existing Customer	40	F	4	High School	Unknown	Less than $40K	Blue	34	3	4	1	3313.0	2517	796.0	1.405	1171	20	2.333	0.760
4	709106358	Existing Customer	40	M	3	Uneducated	Married	$60K - $80K	Blue	21	5	1	0	4716.0	0	4716.0	2.175	816	28	2.500	0.000

About Columns (Features)

1) CLIENTNUM - Client number. Unique identifier for the customer holding the account (ID клиента)

2) Attrition_Flag - Existing and Attrited Customers (Метка принадлежности к оттоку)

3) Customer_Age - Demographic variable - Customer's Age in Years (Возраст)

4) Gender - M = Male, F = Female (Пол)

5) Dependent_count - Number of dependents (Количество детей)

6) Education_Level - Educational Qualification of the account holder (Уровень образования)

(example: high school, college graduate, etc.)

7) Marital_Status - Married, Single, Unknown (Семейное положение)

8) Income_Category - Annual Income Category of the account holder (Категория дохода)

(< $40K, $40K - 60K, $60K - $80K, $80K-$120K, > $120K, Unknown)

9) Card_Category - Type of Card (Blue, Silver, Gold, Platinum) (Категория карты)

10) Months_on_book - Period of relationship with bank (Как давно стал клиентом банка)

11) Total_Relationship_Count - Total no. of products held by the customer ("Количество продуктов банка у кдиента")

12) Months_Inactive_12_mon - Number of months inactive in the last 12 months (Количество месяцев неактивности за последний год)

13) Contacts_Count_12_mon - Number of Contacts in the last 12 months (Количество контактов за последний год)

14) Credit_Limit - Credit Limit on the Credit Card (Кредитный лимит)

15) Total_Revolving_Bal - Total Revolving Balance on the Credit Card (Общий возобновляемый остаток на кредитной карте)

16) Avg_Open_To_Buy - Num Open to Buy Credit Line (Average of last 12 months) ("Количество открытых для покупки кредитной линии за последний год")

17) Total_Amt_Chng_Q4_Q1 - Num Change in Transaction Amount (Q4 over Q1) ("Количественное изменение суммы транзакций относительно 2 периодов")

18) Total_Trans_Amt - Num Total Transaction Amount (Last 12 months) (Общее сумма транзакций)

19) Total_Trans_Ct - Num Total Transaction Count (Last 12 months) (Общее количество транзакций)

20) Total_Ct_Chng_Q4_Q1 - Num Change in Transaction Count (Q4 over Q1) ("Количество изменений в транзакциях относительно 2 периодов")

21) Avg_Utilization_Ratio - Num Average Card Utilization Ratio (Средний коэффициент использования)

Check for duplicate lines (clients).

Проверим, есть ли повторяющиеся строки.

assert data["CLIENTNUM"].nunique() == data.shape[0], "There are duplicates"

data.set_index("CLIENTNUM", inplace = True)

data.head()

	Attrition_Flag	Customer_Age	Gender	Dependent_count	Education_Level	Marital_Status	Income_Category	Card_Category	Months_on_book	Total_Relationship_Count	Months_Inactive_12_mon	Contacts_Count_12_mon	Credit_Limit	Total_Revolving_Bal	Avg_Open_To_Buy	Total_Amt_Chng_Q4_Q1	Total_Trans_Amt	Total_Trans_Ct	Total_Ct_Chng_Q4_Q1	Avg_Utilization_Ratio
CLIENTNUM
768805383	Existing Customer	45	M	3	High School	Married	$60K - $80K	Blue	39	5	1	3	12691.0	777	11914.0	1.335	1144	42	1.625	0.061
818770008	Existing Customer	49	F	5	Graduate	Single	Less than $40K	Blue	44	6	1	2	8256.0	864	7392.0	1.541	1291	33	3.714	0.105
713982108	Existing Customer	51	M	3	Graduate	Married	$80K - $120K	Blue	36	4	1	0	3418.0	0	3418.0	2.594	1887	20	2.333	0.000
769911858	Existing Customer	40	F	4	High School	Unknown	Less than $40K	Blue	34	3	4	1	3313.0	2517	796.0	1.405	1171	20	2.333	0.760
709106358	Existing Customer	40	M	3	Uneducated	Married	$60K - $80K	Blue	21	5	1	0	4716.0	0	4716.0	2.175	816	28	2.500	0.000

data.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 10127 entries, 768805383 to 714337233
Data columns (total 20 columns):
 #   Column                    Non-Null Count  Dtype  
---  ------                    --------------  -----  
 0   Attrition_Flag            10127 non-null  object 
 1   Customer_Age              10127 non-null  int64  
 2   Gender                    10127 non-null  object 
 3   Dependent_count           10127 non-null  int64  
 4   Education_Level           10127 non-null  object 
 5   Marital_Status            10127 non-null  object 
 6   Income_Category           10127 non-null  object 
 7   Card_Category             10127 non-null  object 
 8   Months_on_book            10127 non-null  int64  
 9   Total_Relationship_Count  10127 non-null  int64  
 10  Months_Inactive_12_mon    10127 non-null  int64  
 11  Contacts_Count_12_mon     10127 non-null  int64  
 12  Credit_Limit              10127 non-null  float64
 13  Total_Revolving_Bal       10127 non-null  int64  
 14  Avg_Open_To_Buy           10127 non-null  float64
 15  Total_Amt_Chng_Q4_Q1      10127 non-null  float64
 16  Total_Trans_Amt           10127 non-null  int64  
 17  Total_Trans_Ct            10127 non-null  int64  
 18  Total_Ct_Chng_Q4_Q1       10127 non-null  float64
 19  Avg_Utilization_Ratio     10127 non-null  float64
dtypes: float64(5), int64(9), object(6)
memory usage: 1.6+ MB

Remove from Features "Attrition_Flag" because it's a target.

Не включаю "Attrition_Flag" - так как это целевая переменная.

Features = list(data.columns[1:])
print("Count (Количество) : ", len(Features))
print("Features (Признаки) :\n", Features)

Count (Количество) :  19
Features (Признаки) :
 ['Customer_Age', 'Gender', 'Dependent_count', 'Education_Level', 'Marital_Status', 'Income_Category', 'Card_Category', 'Months_on_book', 'Total_Relationship_Count', 'Months_Inactive_12_mon', 'Contacts_Count_12_mon', 'Credit_Limit', 'Total_Revolving_Bal', 'Avg_Open_To_Buy', 'Total_Amt_Chng_Q4_Q1', 'Total_Trans_Amt', 'Total_Trans_Ct', 'Total_Ct_Chng_Q4_Q1', 'Avg_Utilization_Ratio']

%%time

profile = ProfileReport(data[Features], title = "Summary")

CPU times: user 16.4 s, sys: 1.42 s, total: 17.8 s
Wall time: 16.5 s

#profile.to_file("Obzor.html") # you can save it as html-file
profile.to_notebook_iframe()

data["Attrition_Flag"] = data["Attrition_Flag"].map({"Existing Customer": 0, "Attrited Customer": 1})

X = data[Features]
y = data["Attrition_Flag"]

X.head()

	Customer_Age	Gender	Dependent_count	Education_Level	Marital_Status	Income_Category	Card_Category	Months_on_book	Total_Relationship_Count	Months_Inactive_12_mon	Contacts_Count_12_mon	Credit_Limit	Total_Revolving_Bal	Avg_Open_To_Buy	Total_Amt_Chng_Q4_Q1	Total_Trans_Amt	Total_Trans_Ct	Total_Ct_Chng_Q4_Q1	Avg_Utilization_Ratio
CLIENTNUM
768805383	45	M	3	High School	Married	$60K - $80K	Blue	39	5	1	3	12691.0	777	11914.0	1.335	1144	42	1.625	0.061
818770008	49	F	5	Graduate	Single	Less than $40K	Blue	44	6	1	2	8256.0	864	7392.0	1.541	1291	33	3.714	0.105
713982108	51	M	3	Graduate	Married	$80K - $120K	Blue	36	4	1	0	3418.0	0	3418.0	2.594	1887	20	2.333	0.000
769911858	40	F	4	High School	Unknown	Less than $40K	Blue	34	3	4	1	3313.0	2517	796.0	1.405	1171	20	2.333	0.760
709106358	40	M	3	Uneducated	Married	$60K - $80K	Blue	21	5	1	0	4716.0	0	4716.0	2.175	816	28	2.500	0.000

y.head()

CLIENTNUM
768805383    0
818770008    0
713982108    0
769911858    0
709106358    0
Name: Attrition_Flag, dtype: int64

Num_cols = [cname for cname in Features if data[cname].dtype in ['int64', 'float64']]
Cat_cols = [cname for cname in Features if data[cname].dtype == "object"]

print("Number of Numeric features (Количество числовых признаков) : ", len(Num_cols))
print("Numeric (Числовые признаки) :\n ", Num_cols, " \n")

print("Number of Categorical features (Количество категориальных признаков) : ", len(Cat_cols))
print("Categorical (Категориальные признаки) :\n ", Cat_cols)

Number of Numeric features (Количество числовых признаков) :  14
Numeric (Числовые признаки) :
  ['Customer_Age', 'Dependent_count', 'Months_on_book', 'Total_Relationship_Count', 'Months_Inactive_12_mon', 'Contacts_Count_12_mon', 'Credit_Limit', 'Total_Revolving_Bal', 'Avg_Open_To_Buy', 'Total_Amt_Chng_Q4_Q1', 'Total_Trans_Amt', 'Total_Trans_Ct', 'Total_Ct_Chng_Q4_Q1', 'Avg_Utilization_Ratio']  

Number of Categorical features (Количество категориальных признаков) :  5
Categorical (Категориальные признаки) :
  ['Gender', 'Education_Level', 'Marital_Status', 'Income_Category', 'Card_Category']

y.value_counts()

0    8500
1    1627
Name: Attrition_Flag, dtype: int64

y.value_counts()[0]/y.value_counts()[1]

5.224339274738783

Class Ratio is very different so there is imbalance and I decided to fight with it.

Получается есть Дисбаланс классов.

2) Vizualization (Визуализация)

def build_plot(feature, title_rus):
    trace1 = go.Histogram(
        x = data[data["Attrition_Flag"] == 0][feature],
        name = "Existing Customer (0)",
        marker = dict(color = "blue", opacity = 0.6)
    )
    trace2 = go.Histogram(
        x = data[data["Attrition_Flag"] == 1][feature],
        name = "Attrited Customer (1)",
        marker = dict(color = "red", opacity = 0.8)
    )

    Traces = [trace1, trace2]
    layout = go.Layout(title = title_rus) 
    fig = go.Figure(data = Traces, layout = layout)
    return iplot(fig) 

Watching_Features = ["Income_Category", "Card_Category", "Customer_Age", "Marital_Status", 
                     "Education_Level", "Dependent_count", "Months_on_book",
                     "Credit_Limit", "Total_Trans_Amt", "Total_Trans_Ct"]
Titles_with_Rus = ["Income_Category / Доход", "Card_Category / Тип банковской карты", 
                   "Customer_Age / Возраст клиента", "Marital_Status / Семейное положение", 
                   "Education_Level / Уровень образования", "Dependent_count / Количество детей",
                   "Months_on_book / Как давно стал клиентом банка", "Credit_Limit / Кредитный Лимит",
                   "Total_Trans_Amt / Общая сумма транзакций", "Total_Trans_Ct / Общее количество транзакций"] 

#for a in Watching_Features:
#    build_plot(a, Titles_with_Rus[Watching_Features.index(a)])

build_plot("Income_Category", "Income_Category / Доход")

build_plot("Card_Category", "Card_Category / Тип банковской карты")

build_plot("Customer_Age", "Customer_Age / Возраст клиента")

build_plot("Marital_Status", "Marital_Status / Семейное положение")

build_plot("Education_Level", "Education_Level / Уровень образования")

build_plot("Dependent_count", "Dependent_count / Количество детей")

build_plot("Months_on_book", "Months_on_book / Как давно стал клиентом банка")

build_plot("Credit_Limit", "Credit_Limit / Кредитный Лимит")

build_plot("Total_Trans_Amt", "Total_Trans_Amt / Общая сумма транзакций")

build_plot("Total_Trans_Ct", "Total_Trans_Ct / Общее количество транзакций")

plt.figure(figsize = (12, 8))
plt.title("Зависимость Total_Trans_Ct от Total_Trans_Amt")
sns.scatterplot(data = data, x = "Total_Trans_Amt", y = "Total_Trans_Ct", hue = "Attrition_Flag");

plt.figure(figsize = (12, 8))
plt.title("Зависимость Total_Amt_Chng_Q4_Q1 от Total_Ct_Chng_Q4_Q1")
sns.scatterplot(data = data, x = "Total_Ct_Chng_Q4_Q1", y = "Total_Amt_Chng_Q4_Q1" , hue = "Attrition_Flag");

3) CatBoostClassifier

I know about LightGBMClassifier and XGBoostClassifier (H2OGradientBoostingEstimator, GradientBoostingClassifier and others).

but I prefer CatBoostClassifier! .

Я предпочитаю CatBoostClassifier.

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 1)
print(X_train.shape, y_train.shape, X_test.shape, y_test.shape)

(8101, 19) (8101,) (2026, 19) (2026,)

# But actually, it I may not use StandardScaler() if I use CatBoostClassifier
scaler = StandardScaler()

X_train[Num_cols] = scaler.fit_transform(X_train[Num_cols])
X_test[Num_cols] = scaler.transform(X_test[Num_cols])

# Я не стал вводить здесь Pipeline, так как нет пропущенных значений, 
# не применяю OneHotEncoder - так как применяю CatBoost
# ну и StandardScaler можно и не использовать здесь

params_1 = {'cat_features': Cat_cols,
            'verbose': 1000,
            'random_seed': 1}

model_Cat_1 = CatBoostClassifier(**params_1);
model_Cat_1.fit(X_train, y_train, 
                eval_set = (X_test, y_test),    
                use_best_model = True, 
                plot = True);

Learning rate set to 0.053127
0:	learn: 0.6334559	test: 0.6336110	best: 0.6336110 (0)	total: 65.3ms	remaining: 1m 5s
999:	learn: 0.0123754	test: 0.0641535	best: 0.0638163 (933)	total: 5.63s	remaining: 0us

bestTest = 0.06381629446
bestIteration = 933

Shrink model to first 934 iterations.

It's important for bank for find certainly all client from future churn.

So we will evaluate models by Accuracy and F1 Score, which measure FP and FN cases.

FalsePositive (FP) cases - when we define that these client are from future churn but they are not from Churn.

FalseNegative (FN) cases - when we define that these client are not from future Churn but they are from Churn.

It's import to define FN certainly.

Для банка очень важно точно определить всех клиентов из будущего оттока.

Поэтому будем оценивать модель по Точности (Accuracy) и F1-мере (F1 Score), которая определяет ### ЛП и ЛО случаи.

ЛожноПоложительные (ЛП) случаи - случаи, когда мы определили, что эти клиенты из будущего оттока, а на самом деле нет.

ЛожноНегативные (ЛН) случаи - случаи, когда мы определили. что это клиенты не из будущего оттока, а они из будущего оттока.

Очень важно точно определить ЛН случаи.

%%time

Results = pd.DataFrame(columns = ["model", "roc_auc", "f1_micro"])

scores_r1 = cross_val_score(model_Cat_1, X_test, y_test, 
                            scoring = "roc_auc", 
                            cv = 5,
                            n_jobs = -1)

scores_r2 = cross_val_score(model_Cat_1, X_test, y_test, 
                            scoring = "f1_micro", 
                            cv = 5,
                            n_jobs = -1)

r1 = round(scores_r1.mean(), 6)
r2 = round(scores_r2.mean(), 6)

print("roc_auc", r1)
print("f1_micro", r2)
Results.loc[0] = ["Cat_1", r1, r2]

y_pred = model_Cat_1.predict(X_test)
print(classification_report(y_test, y_pred))

roc_auc 0.988673
f1_micro 0.961506
              precision    recall  f1-score   support

           0       0.98      0.99      0.99      1695
           1       0.94      0.90      0.92       331

    accuracy                           0.97      2026
   macro avg       0.96      0.94      0.95      2026
weighted avg       0.97      0.97      0.97      2026

CPU times: user 69.9 ms, sys: 127 ms, total: 197 ms
Wall time: 12.9 s

params_2 = {'loss_function':'Logloss',
            'eval_metric':'AUC',
            'verbose': 1000,
            'cat_features': Cat_cols,
            'random_seed': 1,}

model_Cat_2 = CatBoostClassifier(**params_2);
model_Cat_2.fit(X_train, y_train, 
                eval_set = (X_test, y_test),    
                use_best_model = True, 
                plot = True);

Learning rate set to 0.053127
0:	test: 0.8826110	best: 0.8826110 (0)	total: 6.7ms	remaining: 6.69s
999:	test: 0.9948845	best: 0.9949968 (350)	total: 5.71s	remaining: 0us

bestTest = 0.9949968363
bestIteration = 350

Shrink model to first 351 iterations.

%%time

scores_r1 = cross_val_score(model_Cat_2, X_test, y_test, 
                            scoring = "roc_auc", 
                            cv = 5,
                            n_jobs = -1)

scores_r2 = cross_val_score(model_Cat_2, X_test, y_test, 
                            scoring = "f1_micro", 
                            cv = 5,
                            n_jobs = -1)

r1 = round(scores_r1.mean(), 6)
r2 = round(scores_r2.mean(), 6)

print("roc_auc", r1)
print("f1_micro", r2)
Results.loc[1] = ["Cat_2", r1, r2]

y_pred = model_Cat_2.predict(X_test)
print(classification_report(y_test, y_pred))

roc_auc 0.988673
f1_micro 0.961506
              precision    recall  f1-score   support

           0       0.98      0.99      0.99      1695
           1       0.95      0.90      0.92       331

    accuracy                           0.98      2026
   macro avg       0.96      0.95      0.95      2026
weighted avg       0.98      0.98      0.98      2026

CPU times: user 31.7 ms, sys: 4.8 ms, total: 36.5 ms
Wall time: 10.7 s

I will try SomteNC library.

Решил попробовать Стратегию Smote (SmoteNC - с учётом категориальных данных)!!!

More information about imbalance https://dyakonov.org/2021/05/27/imbalance/

Подробнее о дисбалансе: https://dyakonov.org/2021/05/27/imbalance/

#from imblearn.over_sampling import SMOTE
from imblearn.over_sampling import SMOTENC

SMOTE - library to work with imbalance and all Features must be in Numeric format.

SMOTENC - library to work with imbalance and you can work wirh Categorical Features.

SMOTE - библиотека для работы с дисбалансом и все признаки должны быть переведены в цифру.

SMOTENC - библиотека для работы с дисбалансом и признаки могут быть категориальными.

Cat_cols

['Gender',
 'Education_Level',
 'Marital_Status',
 'Income_Category',
 'Card_Category']

index_of_Cat = []

listik = list(X.columns)
for a in Cat_cols:
    index_of_Cat.append(listik.index(a))
print("Count (Количество) : ", len(index_of_Cat))
print("Index of Categorical (Индексы категориальных признаков) : ", index_of_Cat)

Count (Количество) :  5
Index of Categorical (Индексы категориальных признаков) :  [1, 3, 4, 5, 6]

smoteNC = SMOTENC(random_state=42, categorical_features = index_of_Cat)

X_smoteNC, y_smoteNC = smoteNC.fit_resample(X, y) 

print(f'''Shape of X before SMOTENC (Размер X до): {X.shape}
Shape of X after SMOTENC (Размер X после): {X_smoteNC.shape}''')

print('\nBalance of positive and negative classes (%): (Соотношение положительных и негативных случаев)')
y_smoteNC.value_counts(normalize=True) * 100

Shape of X before SMOTENC (Размер X до): (10127, 19)
Shape of X after SMOTENC (Размер X после): (17000, 19)

Balance of positive and negative classes (%): (Соотношение положительных и негативных случаев)

0    50.0
1    50.0
Name: Attrition_Flag, dtype: float64

X_train_smNC, X_test_smNC, y_train_smNC, y_test_smNC = train_test_split(X_smoteNC, y_smoteNC, test_size = 0.2, random_state = 1)
print(X_train_smNC.shape, y_train_smNC.shape, X_test_smNC.shape, y_test_smNC.shape)

(13600, 19) (13600,) (3400, 19) (3400,)

scaler = StandardScaler()

X_train_smNC[Num_cols] = scaler.fit_transform(X_train_smNC[Num_cols])
X_test_smNC[Num_cols] = scaler.transform(X_test_smNC[Num_cols])

params_3 = {'cat_features': Cat_cols,
            'verbose': 1000,
            'random_seed': 1,}

model_Cat_3 = CatBoostClassifier(**params_3);
model_Cat_3.fit(X_train_smNC, y_train_smNC, 
                eval_set = (X_test_smNC, y_test_smNC),    
                use_best_model = True, 
                plot = True);

Learning rate set to 0.06038
0:	learn: 0.6174961	test: 0.6175708	best: 0.6175708 (0)	total: 10.5ms	remaining: 10.5s
999:	learn: 0.0133494	test: 0.0437913	best: 0.0437758 (993)	total: 8.77s	remaining: 0us

bestTest = 0.04377581631
bestIteration = 993

Shrink model to first 994 iterations.

%%time

scores_r1 = cross_val_score(model_Cat_3, X_test_smNC, y_test_smNC, 
                            scoring = "roc_auc", 
                            cv = 5,
                            n_jobs = -1)

scores_r2 = cross_val_score(model_Cat_3, X_test_smNC, y_test_smNC, 
                            scoring = "f1_micro", 
                            cv = 5,
                            n_jobs = -1)

r1 = round(scores_r1.mean(), 6)
r2 = round(scores_r2.mean(), 6)

print("roc_auc", r1)
print("f1_micro", r2)
Results.loc[2] = ["Cat_3", r1, r2]

y_pred = model_Cat_3.predict(X_test_smNC)
print(classification_report(y_test_smNC, y_pred))

roc_auc 0.995795
f1_micro 0.972647
              precision    recall  f1-score   support

           0       0.99      0.98      0.98      1678
           1       0.98      0.99      0.99      1722

    accuracy                           0.98      3400
   macro avg       0.99      0.98      0.98      3400
weighted avg       0.99      0.98      0.98      3400

CPU times: user 41.8 ms, sys: 6.02 ms, total: 47.8 ms
Wall time: 13.2 s

Results.set_index("model", inplace = True)
Results

	roc_auc	f1_micro
model
Cat_1	0.988673	0.961506
Cat_2	0.988673	0.961506
Cat_3	0.995795	0.972647

def plot_Importance(model):
    Feat_Importance = {"Names": Features, "Importance": model.get_feature_importance()}
    Import_Data = pd.DataFrame(Feat_Importance)
    Import_Data.sort_values(by = ["Importance"], ascending = False, inplace = True) 
    plt.figure(figsize = (15, 10))
    
    sns.barplot(x = Import_Data["Importance"], y = Import_Data["Names"])
    plt.title("Feature Importance / Важность признаков")
    plt.xlabel("Importance / Важность")
    plt.ylabel("Names / Признак")

plot_Importance(model_Cat_3)

4) Economic effect (Экономический эффект)

data.head()

	Attrition_Flag	Customer_Age	Gender	Dependent_count	Education_Level	Marital_Status	Income_Category	Card_Category	Months_on_book	Total_Relationship_Count	Months_Inactive_12_mon	Contacts_Count_12_mon	Credit_Limit	Total_Revolving_Bal	Avg_Open_To_Buy	Total_Amt_Chng_Q4_Q1	Total_Trans_Amt	Total_Trans_Ct	Total_Ct_Chng_Q4_Q1	Avg_Utilization_Ratio
CLIENTNUM
768805383	0	45	M	3	High School	Married	$60K - $80K	Blue	39	5	1	3	12691.0	777	11914.0	1.335	1144	42	1.625	0.061
818770008	0	49	F	5	Graduate	Single	Less than $40K	Blue	44	6	1	2	8256.0	864	7392.0	1.541	1291	33	3.714	0.105
713982108	0	51	M	3	Graduate	Married	$80K - $120K	Blue	36	4	1	0	3418.0	0	3418.0	2.594	1887	20	2.333	0.000
769911858	0	40	F	4	High School	Unknown	Less than $40K	Blue	34	3	4	1	3313.0	2517	796.0	1.405	1171	20	2.333	0.760
709106358	0	40	M	3	Uneducated	Married	$60K - $80K	Blue	21	5	1	0	4716.0	0	4716.0	2.175	816	28	2.500	0.000

data[data["Attrition_Flag"] == 1]["Income_Category"].value_counts()

Less than $40K    612
$40K - $60K       271
$80K - $120K      242
$60K - $80K       189
Unknown           187
$120K +           126
Name: Income_Category, dtype: int64

Money_Churn = data[data["Attrition_Flag"] == 1]["Income_Category"].value_counts()

Money_Churn[0]

612

I decided to calculate how much we lose in money,

starting from the mean of the given money category and

how many cases we have.

Решил посчитать сколько теряем в деньгах,

отталкиваясь от среднего данной денежной категории и

количества таких случаев в этой категории.

Suppose, that the bank makes in average 8-10% profit from one account.

So profit_koef = 0.08.

Предположим, что банк имеет 8-10 : прибыли с каждого счёта

Поэтому коэффициент_прибыли = 0.08 (profit_koef).

profit_koef = 0.08
Summa = profit_koef*(Money_Churn[0]*20000 + Money_Churn[1]*50000 + Money_Churn[2]*100000 + Money_Churn[3]*70000 + \
        Money_Churn[5]*120000 + Money_Churn[4]*20000)
print("Summa : ", Summa)

Summa :  6566400.0

from IPython.display import Image
Image(filename = "Carl_Meme.jpg")

Strategy: allocate some money to retain (some part of the future churn).

Bank can find out the reasons for the churn and offer more comfortable service!

Стратегия борьбы : выделить некоторые средства для удержания (некоторой части из будущего оттока).

путём выяснения причин оттока и предложения тарифа на более выгодных условиях!

Thanks for reading.

Спасибо за внимание.